Abstract: Graph Convolutional Networks (GCNs) have demonstrated significant potential in graph representation learning. However, existing methods still exhibit limitations in learning global topological relationships and effectively integrating topological structures with attribute features. To address these challenges, this paper proposes a Graph Convolutional Network Enhanced by Graph Diffusion and Dual-View Feature Learning (GCN-GDDV). The GCN-GDDV framework first introduces a generalized graph diffusion mechanism to construct diffusion graphs that encapsulate global topological information. It then combines these diffusion graphs with attribute feature-based K-nearest neighbor (KNN) graphs to perform dual-view feature learning via graph convolutional networks. This dual-view approach separately captures global structural dependencies and node attribute semantic similarities. Finally, an attention network is designed to adaptively fuse topological structures and attribute features. Node classification experiments conducted on three benchmark graph datasets demonstrate that GCN-GDDV outperforms state-of-the-art baselines, achieving improvements of 2.7%, 1.5%, and 1.0% in Accuracy, Macro-F1, and Micro-F1 metrics, respectively.

Key words: graph convolutional network, graph diffusion, dual-view feature learning, attention mechanism, node classification

摘要： 图卷积网络在图表示学习领域已展现了强大的潜力，然而现有方法在全局拓扑关系学习、拓扑结构和属性特征融合方面仍存在局限性。针对该问题，提出一种图扩散与双视图特征学习增强的图卷积网络(Graph Convolutional Network Enhanced by Graph Diffusion and Dual-View Feature Learning，GCN-GDDV)。GCN-GDDV首先引入广义图扩散机制构建包含全局拓扑结构信息的扩散图，然后结合属性特征K近邻图，进行基于图卷积网络的双视图特征学习，以分别捕捉全局结构关系依赖与节点属性的语义相似性，最后设计注意力网络实现拓扑结构和属性特征的自适应融合。在三个常用图数据集上进行节点分类验证实验，结果表明GCN-GDDV相对于最优的基线在Accuracy、Macro-F1和Micro-F1指标上分别提升2.7%、1.5%和1.0%。

关键词: 图卷积网络, 图扩散, 双视图特征学习, 注意力机制, 节点分类